Tension control method for cling film winding process

Abstract

This invention relates to the field of plastic wrap winding technology, and more particularly to a tension control method for the plastic wrap winding process. By acquiring standard image sequences of flattened plastic wrap and extracting quantitative features, a tension mapping model based on reinforcement learning is constructed, and finally, an adjustment command for the overall pressure roller is output. This achieves the purpose of sensing the microscopic physical state of the plastic wrap surface and adjusting the pressure roller, significantly improving the internal quality of the film roll. By adopting an optical imaging scheme that combines a low-angle grazing light source with a light-absorbing background, and subsequently using illumination homogenization correction based on polynomial surface fitting, the imaging contrast of microscopic wrinkles on the transparent film surface is greatly enhanced, and the interference of uneven illumination in the industrial environment is eliminated to a certain extent. This provides high-quality image data for the subsequent calculation of the first, second, and third quantitative features, thereby improving the accuracy of the final calculated pressure roller adjustment amount.

Description

Technical Field

[0001] This invention relates to the field of plastic wrap winding technology, and in particular to a tension control method for the plastic wrap winding process. Background Technology

[0002] In the automated winding production of cling film, maintaining constant winding tension is the core link to ensure the quality of the film roll. Most existing technologies adopt tension control methods based on tension sensors. For example, by detecting the real-time tension of the film between the guide rollers, comparing it with the preset target tension value, and using control algorithms such as PID to drive actuators such as magnetic powder brakes and servo motors for adjustment.

[0003] However, for ultra-thin or wide polyethylene (PE) or polyvinyl chloride (PVC) cling film, the aforementioned tension control method occasionally encounters the following problems during use: it struggles to detect and address localized micro-wrinkles on the film surface caused by minute parallelism errors in the guide rollers, uneven roller surface temperature, or environmental airflow disturbances. During the high-speed winding of the cling film, although some local wrinkles may be stretched and smoothed to some extent due to the extensibility of the cling film material, keeping the tension sensor's detection value within the set range, uneven stress distribution and micro-wrinkles have already formed inside the film. These uncorrected micro-wrinkles are layered and compressed during the winding process, eventually leading to hard spots or chrysanthemum patterns inside the film roll. This can easily cause quality problems such as film roll breakage, edge tearing, or uneven film exit during subsequent slitting or user stretching. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a tension control method for the plastic wrap winding process, which solves the technical problem that existing technologies struggle to detect and adjust the microscopic physical state of the plastic wrap surface, leading to abnormal quality in the plastic wrap.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a tension control method for the plastic wrap winding process, applied to the pressure roller for tension control in the plastic wrap winding process, comprising the following steps: S1. Collect the original film image sequence in the flat state before the plastic wrap is rolled up, and perform preprocessing to obtain a standardized image sequence containing multiple standardized images collected periodically. S2. Based on multiple standardized image sequences, calculate the first quantitative feature value in each standardized image that reflects the degree of wrinkling on the membrane surface, the second quantitative feature value that reflects the degree of membrane surface deviation, and the third quantitative feature value that reflects the trend of membrane surface deviation. S3. Input the first quantitative feature value, the second quantitative feature value, the third quantitative feature value and the real-time process parameters of the winding process into a pre-established tension mapping model for characterizing the degree of film wrinkling, the degree of film deviation, and the relationship between process parameters and pressure roller adjustment amount. The tension mapping model is constructed based on a reinforcement learning model. The pressure roller adjustment amount includes the pressure adjustment amount, tilt angle adjustment amount and lateral displacement adjustment amount of the pressure roller. S4. Adjust the pressure roller parameters according to the pressure roller adjustment amount.

[0006] Preferably, the specific steps for acquiring the original film image sequence in the flattened state before the cling film is rolled up are as follows: S111. When light from the light source shines on the surface of the plastic wrap, the angle between the incident direction of the light and the normal to the surface of the plastic wrap is greater than 75 degrees. S112. A light-absorbing background is provided on the back of the plastic wrap, wherein the light-absorbing background is a black rough surface material with a spectral reflectance of less than 5%. S113. An image acquisition device is set up to capture images of the plastic wrap in a flat state before it is rolled up, and the angle between the shooting direction of the image acquisition device and the illumination direction of the light source is in the range of [15°, 45°]. S114. Images of the plastic wrap at various times are acquired in real time using an image acquisition device and arranged in chronological order to obtain the original film image sequence.

[0007] Preferably, in step S1, the specific steps of the preprocessing are as follows: S121. Filter each original thin film image in the original thin film image sequence to obtain a filtered image; S122. Construct a background function model to characterize the ideal background grayscale value of each pixel in the filtered image. Its expression is: In the above formula, Indicates the coordinates in the filtered image as The ideal background grayscale value for the pixel at point n, where n is the order of the polynomial function. For coefficients; S123. Grid the filtered image and calculate the standard deviation of the gray values ​​of each pixel in each grid in turn; S124. Select multiple grids with the smallest standard deviation as candidate grids; S125. Calculate the average gray value of all pixels in each candidate grid in turn, and select the pixel corresponding to the gray value that is closest to the average value as the candidate pixel corresponding to the candidate grid. S126. By fitting the background function model using the least squares method, the sum of squared errors between the gray values ​​of the background function model and the true gray values ​​at all candidate pixels is minimized in order to solve for the coefficients and obtain the background function model. S127. Subtract the background function model from the filtered image and adjust the grayscale values ​​to obtain the corrected uniform illumination image. The expression is: In the above formula, Represents coordinates in a uniformly lit image The pixel value at that pixel. Indicates the coordinates in the filtered image as The pixel value at the pixel point, and C represents the adjustment coefficient used to adjust the pixel value of each pixel in the uniformly illuminated image to a suitable level; S128. Linearly contrast stretch the uniformly illuminated image to make its grayscale value range reach a preset range, so as to obtain a standardized image and a standardized image sequence composed of multiple standardized images arranged in time order.

[0008] Preferably, in step S2, the specific steps for calculating the first quantized feature value are as follows: S211. The standardized image is meshed to obtain multiple non-overlapping rectangular units; S212. Calculate the grayscale gradient vector of each pixel within each rectangular unit. The expression is as follows: In the above formula, Represents pixels The gray gradient vector at that point, and Representing pixels The gradient components of the gray-level gradient vector at a given location in the x and y directions; S213. Normalize the gray-level gradient vector of each pixel in the rectangular unit, and calculate the average direction vector that represents the main direction of the gray-level gradient vector corresponding to each pixel in the rectangular unit. S214. Calculate the mean value of the projection of the gray-level gradient vector of each pixel within the rectangular unit onto the average direction vector to obtain the gradient consistency index. The calculation formula is as follows: In the above formula, This represents the gradient consistency index of the rectangular cell located in the m-th row and n-th column after meshing. This represents the average direction vector of the rectangular cell located in the m-th row and n-th column after meshing. This represents the set of all pixels located in the rectangular cell at row m and column n after meshing. Indicates the size of the set. The L2 norm of a vector; S215. Establish multiple parallel grayscale profile lines on the plane where the rectangular unit is located, all of which are perpendicular to the grayscale gradient vector, and obtain the grayscale value corresponding to each pixel on the grayscale profile line to obtain the grayscale sequence. S216. Based on the total number of peaks and troughs corresponding to all grayscale sequences in the rectangular unit, calculate the average of the total number of peaks and troughs contained in each grayscale profile line to obtain the grayscale volatility index. The calculation formula is as follows: In the above formula, This represents the grayscale fluctuation index of the rectangular cell located in the m-th row and n-th column after gridding. This represents the total number of peaks and troughs in the grayscale sequence corresponding to the k-th grayscale profile line. grayscale profile lines; S217. Calculate the local wrinkling index of each rectangular unit based on its gradient consistency index and grayscale fluctuation index. The calculation formula is as follows: In the above formula, This represents the local wrinkling index of the rectangular cell located in the m-th row and n-th column after meshing; S218. Divide the plastic wrap into multiple non-overlapping statistical units along the direction of movement of the plastic wrap, and calculate the ratio of the number of rectangular units in each statistical unit whose local wrinkling index is greater than the preset wrinkling index threshold to the total number of rectangular units in order to obtain the instantaneous wrinkling index of each statistical unit. S219. Calculate the average value of the instantaneous wrinkling index of each statistical unit within a certain period of time from the current time, so as to obtain the first quantitative characteristic value of the statistical unit at the current time.

[0009] Preferably, in step S213, the calculation of the average direction vector is as follows: S2131. Calculate and double the orientation angle of each pixel within the rectangular unit based on its grayscale gradient vector. The calculation formula is as follows: In the above formula, Represents the coordinates within a rectangular cell The direction angle of the grayscale gradient vector corresponding to the pixel at that location. It is the angle after doubling the direction angle. and Representing pixels The gradient components of the gray-level gradient vector at a given location in the x and y directions; S2132. The normalized grayscale gradient vectors of the pixels corresponding to the doubled direction angle are synthesized to obtain a synthesized vector. The formula for calculating the synthesized vector is: In the above formula, and These represent the components of the composite vector in the x and y directions, respectively. S2133. Calculate the direction angle of the composite vector and divide it by 2 to obtain the true direction angle of the composite vector. S2134. Calculate the average direction vector based on the true direction angle of the composite vector.

[0010] Preferably, the specific steps for calculating the second and third quantized feature values ​​are as follows: S221. Establish two pre-calibrated parallel theoretical baselines on the standardized image to reflect the position of the edge of the plastic wrap. S222. Perform edge detection on the standardized image in real time to identify the actual pixel points on both sides of the plastic wrap. S223. Fit the actual pixel points on both sides of the plastic wrap to obtain the real-time edge lines on both sides of the plastic wrap. S224. Establish multiple measurement lines on the standardized image that are perpendicular to and parallel to the two theoretical baselines, and obtain the intersection points of each measurement line with the theoretical baseline and the real-time edge line, and mark them as theoretical intersection points and real-time intersection points respectively. S225. Based on the coordinates of the theoretical intersection point and the real-time intersection point, calculate the Euclidean distance between the theoretical intersection point and the real-time intersection point on the same side edge of the plastic wrap on each measurement line in turn to obtain the deviation on one side. When the real-time intersection point is located on one side of the theoretical baseline, mark the Euclidean distance as positive and when the real-time intersection point is located on the other side of the theoretical baseline, mark the Euclidean distance as negative. S226. Calculate the average value of the deviation on the two sides corresponding to the measurement line to obtain the local deviation. S227. Calculate the average value of the local deviation corresponding to each measurement line to obtain the instantaneous deviation, and mark it as the second quantization feature value. Arrange the instantaneous deviation calculated at each time point in chronological order to obtain the instantaneous deviation sequence. S228. Set a statistical window for a period of time from the current time, and perform linear regression analysis on the data of the instantaneous deviation sequence within the statistical window to calculate its regression slope and obtain the third quantitative feature value.

[0011] Preferably, in step S3, the tension mapping model is constructed based on a reinforcement learning model, and the specific steps are as follows: A state space is constructed, which includes the second and third quantized feature values ​​at the current moment, the first quantized feature values ​​corresponding to each statistical unit, the current pressure of the pressure roller, the current tilt angle of the pressure roller, the current lateral displacement of the pressure roller, and the real-time process parameters of the winding process. Construct an action space, which includes three pressure roller adjustment amounts: pressure adjustment amount, tilt angle adjustment amount, and lateral displacement adjustment amount. Construct a reward function, which is calculated based on the first quantized feature value, the second quantized feature value, and the third quantized feature value at the next time step and the current time step; Based on the state space, action space, and reward function, a deep reinforcement learning algorithm is used to train the tension mapping model.

[0012] Preferably, the reward function is a weighted sum of wrinkle improvement reward, deviation correction reward, tension stability reward, and safety constraint penalty. The wrinkle improvement reward is calculated by summing the reduction of the first quantitative feature value corresponding to each statistical unit compared to the current time, and is positively correlated with the sum of the reductions. The deviation correction reward is calculated by comparing the absolute values ​​of the second and third quantitative feature values ​​at the next time step with the current time step, and is positively correlated with the absolute values ​​of the second and third quantitative feature values. The tension stability bonus is calculated by normalizing the current pressure adjustment, the tilt angle adjustment of the pressure roller, the lateral displacement adjustment of the pressure roller, and the maximum allowable adjustment range of each parameter, and then calculating the sum of squares of the normalized parameters and negatively correlated with this sum of squares. The safety constraint penalty is set to a value that is much greater than the preset values ​​of the wrinkle improvement reward, deviation correction reward, and tension stability reward when the tension of the adjusted plastic wrap is greater than its allowable safety tension threshold; otherwise, it is 0.

[0013] Preferably, in step S2134, the formula for calculating the average direction vector is: In the above formula, Represents the average direction vector. This represents the true direction angle of the composite vector.

[0014] The present invention also provides a tension control system for the plastic wrap winding process, including a processor and a memory, wherein the memory is used to store a computer program, and the computer program is executed by the processor to implement the tension control method.

[0015] By employing the above technical solution, the present invention provides a tension control method for the plastic wrap winding process, which has at least the following beneficial effects: 1. This invention acquires standard image sequences of flattened cling film and extracts quantitative features to construct a tension mapping model based on reinforcement learning. Finally, it outputs adjustment commands for the overall pressure roller, thereby achieving the purpose of sensing the microscopic physical state of the cling film surface and adjusting the pressure roller. This solves the inherent defects of traditional tension control systems that are difficult to sense and respond to microscopic wrinkles and deviations on the film surface in real time, and significantly improves the internal quality of the film roll and its cutting and use performance.

[0016] 2. This invention employs an optical imaging scheme combining a low-angle grazing light source and a light-absorbing background, and subsequently uses illumination homogenization correction based on polynomial surface fitting. This greatly enhances the imaging contrast of micro-wrinkles on the transparent film surface, and to a certain extent eliminates the interference of uneven illumination in industrial settings, providing reliable high-quality images for subsequent processing.

[0017] 3. This invention uses a method to calculate the local wrinkling index by fusing gradient direction consistency and grayscale profile fluctuation. This method integrates two features that characterize different physical properties of wrinkles, making the algorithm highly sensitive to periodic directional wrinkles and robust to interference such as scratches, stains, and noise. This enables accurate and stable quantitative evaluation of micro-wrinkles.

[0018] 4. This invention designs a reinforcement learning decision framework with state space, action space and reward function as the core, which enables the agent to learn to optimize the film surface morphology globally by coordinating the adjustment of pressure, tilt angle and displacement under the physical constraint of a single pressure roller. To a certain extent, it realizes the dynamic balance and intelligent decision-making of multiple conflicting objectives such as wrinkle elimination, deviation correction, control smoothness and safety assurance.

[0019] 5. This invention achieves automatic and real-time precise control of winding tension through automatic visual inspection, intelligent decision-making and execution closed loop. It avoids the problems of relying on employee experience and frequent machine stoppages for manual adjustment in traditional production. It significantly reduces quality abnormalities such as film breakage and wrinkling caused by improper tension and subsequent downtime, thereby ensuring the continuous and stable operation of the production line and improving the overall efficiency and capacity of the equipment. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the tension control method for the plastic wrap winding process according to the present invention; Figure 2This is a schematic diagram showing the division of different areas on the cling film of the present invention; Figure 3 Microscopic images of thin film wrinkles. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0022] To address the technical problem of existing technologies' inability to detect and adjust localized micro-wrinkles on the surface of cling film, leading to quality defects, this invention provides a tension control method for the cling film winding process. Applied to the pressure roller in the cling film winding process, this method can identify the micro-state of the cling film surface and control the tension in real time, thereby reducing the probability of quality problems in the cling film. Figure 1 As shown, the tension control method includes the following steps: S1. First, it is necessary to acquire image data that clearly and accurately reflects the physical state of the plastic wrap surface before winding. However, since microscopic wrinkles and plastic wrap misalignment cannot be directly observed after it is wound, and considering that the tension disturbance during the winding process will be reflected in the surface morphology of the film in real time, it is necessary to acquire images at a key position when the plastic wrap is in a flattened state and about to be wound. Generally, image acquisition is carried out between the last guide roller and the winding roller, that is, to acquire the original film image sequence in the flattened state before the plastic wrap is wound. The original film image sequence consists of multiple original film images arranged in chronological order. Since the microscopic features of the plastic wrap surface are relatively subtle and difficult to capture, the following provides a specific method for acquiring the original film image sequence in the flattened state before the plastic wrap is wound, in order to improve the image acquisition effect of the physical features of the plastic wrap surface: S111. Considering that the surface of plastic wrap is flat, tiny folds with a height difference of only a few micrometers to tens of micrometers may not produce much contrast under normal front lighting, making them difficult to image. Therefore, a low-angle grazing illumination method is adopted, so that when the light source shines on the surface of the plastic wrap, the angle between the incident direction of the light and the normal of the surface of the plastic wrap is greater than 75 degrees. This creates a certain shadow on the back of the tiny folds, while the light-facing side of the tiny folds forms a highlight. This converts the weak vertical height difference into horizontal grayscale contrast in the image, enabling subsequent processing methods to effectively detect these features.

[0023] S112. Considering that plastic wrap usually has high or semi-transparent light transmission, light will penetrate the plastic wrap and be reflected back by the mechanical structure behind it, which may form a relatively complex background interference in the acquired image, thus potentially obscuring the edge and texture information of the plastic wrap itself. Therefore, a light-absorbing background is set on the back of the plastic wrap. The light-absorbing background is a black rough surface material with a spectral reflectance of less than 5%. This light-absorbing background can absorb most of the incident light, making the background area in the image close to pure black, thereby maximizing the contrast between the plastic wrap as the foreground and the background, improving the clarity of the edge of the plastic wrap, and providing a basis for subsequent accurate edge detection and texture analysis.

[0024] S113. After determining the incident angle of the light source and the light-absorbing background, the shooting direction of the image acquisition device also needs to avoid the specular reflection angle. This is because if the image acquisition device is directly facing the reflected light path, the smooth surface of the plastic wrap will directly reflect the strong light into the lens of the image acquisition device, resulting in local overexposure and loss of all details in that area. Therefore, an image acquisition device is set up to capture images of the plastic wrap in its flat state before it is rolled up, and the angle between the shooting direction of the image acquisition device and the illumination direction of the light source is in the range of [15°, 45°]. [15°, 45°] is an empirically optimized range. This angle ensures that the image acquisition device receives information such as scattered light from the surface of the film and shadows and highlights generated by wrinkles, rather than direct specular reflection light, thereby obtaining images with richer details.

[0025] S114. Real-time images of the plastic wrap are acquired using an image acquisition device and arranged in chronological order to obtain the original film image sequence, which facilitates better subsequent analysis of the changes in the first quantization feature value, the second quantization feature value, and the third quantization feature value.

[0026] After obtaining the original film image sequence, preprocessing is required to obtain a standardized image sequence containing multiple standardized images acquired periodically. The standardized image is the preprocessed version of the original film image. The purpose of preprocessing is to eliminate, to some extent, the effects of uneven lighting, noise, and equipment interference in the original film image. This is because during the production process, obstructions from equipment and other objects in the factory can cause variations in the brightness of the light illuminating the film, and the operation of various equipment in the factory can interfere with the image acquisition device, potentially resulting in noise. Therefore, preprocessing provides a more accurate data foundation for subsequent quantitative analysis. A detailed preprocessing method is provided below: S121. The image acquisition device and its circuit may introduce high-frequency random noise into the acquired image. Therefore, it is necessary to perform adaptive median filtering or Gaussian filtering on each original thin film image in the original thin film image sequence to obtain a filtered image.

[0027] S122. Illumination unevenness is a low-frequency signal with slow spatial variation. Therefore, a background function model is constructed to characterize the ideal background gray value of each pixel in the filtered image. This background function model adopts a bivariate polynomial function, which is flexible in form and can fit various common illumination distributions, such as plane tilt and central spot, and is easy to solve using the least squares method. Its expression is: In the above formula, Indicates the coordinates in the filtered image as The ideal background grayscale value for a pixel is defined by n, where n is the order of the polynomial function, typically 2 or 3. A low order may lead to underfitting, while a high order may fit the actual wrinkle features into the background, resulting in information loss after correction. The coefficients determine the surface shape of the established background function model.

[0028] Next, the background function model needs to be solved. Before fitting, the pixels with the highest probability of being background need to be obtained from the filtered image. The specific method is as follows: S123. Considering that the local gray values ​​of the background area are relatively uniform, the filtered image is gridded, and the standard deviation of the gray value of each pixel in each grid is calculated in turn. The smaller the standard deviation, the more uniform the gray value of the pixel in the grid is, and the more likely it is to be the background area. Here, the background area refers to a flat film surface without wrinkles or other film surface features.

[0029] S124. Select multiple grids with the smallest standard deviation as candidate grids.

[0030] S125. Calculate the average gray value of all pixels in each candidate grid in turn, and select the pixel corresponding to the gray value that is closest to the average value as the candidate pixel corresponding to the candidate grid. This is to take into account that there may still be small fluctuations at the pixel level within each candidate grid. Simply taking the center point of the grid or a random point may just select an outlier point that deviates from the overall level. By calculating the average value, the overall trend of the gray value of each pixel in the candidate grid is reflected. Selecting the pixel whose gray value is closest to the average value can, to a certain extent, ensure that the selected pixel can best represent the overall gray value level of the candidate grid.

[0031] S126. By fitting the data using the least squares method, the sum of the squared errors between the gray values ​​of the background function model and the true gray values ​​at all candidate pixels is minimized. The coefficients are then solved to obtain the background function model.

[0032] S127. Subtract the background function model from the filtered image and adjust the grayscale values ​​to obtain the corrected uniform illumination image. The expression is: In the above formula, Represents coordinates in a uniformly lit image The pixel value at that pixel. Indicates the coordinates in the filtered image as The pixel value at the point is C, which represents the adjustment coefficient used to adjust the pixel value of each pixel in the uniformly illuminated image to a suitable level. After this step, the gray value of each pixel in the uniformly illuminated image mainly reflects the reflective properties, transmission properties and surface texture of the film itself, while the large gray gradient caused by illumination has been basically eliminated. At this time, wrinkles of the same depth should show similar gray values ​​at any image position.

[0033] S128. Considering that the gray value range of pixels in a uniformly illuminated image is small, the uniformly illuminated image is linearly contrast stretched to make its gray value range reach a preset range, so as to obtain a standardized image and a standardized image sequence composed of multiple standardized images arranged in time order.

[0034] After the preprocessing step, information directly related to winding quality and used for subsequent calculations is enhanced and standardized, such as information about wrinkles and edges, while interference information unrelated to winding quality, such as information about lighting and noise, is greatly suppressed.

[0035] S2. The defects of the plastic wrap that this embodiment aims to improve are mainly reflected in two aspects: the smoothness of the plastic wrap surface and the accuracy of the movement trajectory. First, based on multiple standardized image sequences, the first quantitative feature value used to reflect the degree of wrinkling on the film surface is calculated in each standardized image. Considering that wrinkles appear as a texture with a specific direction and periodic alternation of light and dark on the image, they are analyzed and calculated using the following method: S211. Minor wrinkles usually appear locally on the surface of plastic wrap. Therefore, in order to more accurately analyze the local characteristics of different areas on the surface of plastic wrap, such as... Figure 2 As shown, the normalized image is meshed to obtain multiple non-overlapping rectangular units, i.e. Figure 2 The dashed rectangle in the image is used to locate the specific area where the wrinkles appear.

[0036] S212. For wrinkles, the edges of their light and dark stripes exhibit strong grayscale changes. The grayscale gradient vector is an effective tool for describing the direction and intensity of local grayscale changes in an image. Therefore, the expression for calculating the grayscale gradient vector of each pixel within each rectangular unit is as follows: In the above formula, Represents pixels The gray gradient vector at that point, and Representing pixels The gradient components of the gray-level gradient vector at a given point in the x and y directions, that is, the rate of change of the gray-level value at that pixel along the x and y axes, can be calculated using gradient operators such as Sobel and Prewitt.

[0037] S213. If there are obvious directional textures within a region, such as folds in a single direction, etc. Figure 3 Given the thin film microstructure shown, the gradient directions of most pixels within this region should be roughly the same. Therefore, the gray-level gradient vector of each pixel within the rectangular unit is normalized, and the average direction vector, which characterizes the dominant direction of the gray-level gradient vector corresponding to each pixel within the rectangular unit, is calculated. This average direction vector represents the dominant orientation of the texture within the rectangular unit, which could be wrinkles, impurities, scratches, etc. However, considering that directly calculating the average direction vector using the arithmetic mean would fail due to cyclic orientation (e.g., 0° and 360° represent the same direction, but the values ​​differ greatly) and opposite direction cancellation (the gradient directions of the bright and dark edges of a wrinkle are opposite), the following method for calculating the average direction vector is provided: S2131. Calculate and double the orientation angle of each pixel within the rectangular unit based on its grayscale gradient vector. The calculation formula is as follows: In the above formula, Represents the coordinates within a rectangular cell The direction angle of the grayscale gradient vector corresponding to the pixel at that location. This is the angle after doubling the direction angle. Doubling makes two vectors with opposite directions (i.e., two vectors that are 180° apart) become vectors with the same direction (i.e., vectors that are 360° apart or 0° apart) after doubling. Therefore, they will not cancel each other out when averaging, thus correctly reflecting the texture direction. and Representing pixels The gradient components of the gray-level gradient vector at a given location in the x and y directions.

[0038] S2132. The normalized grayscale gradient vectors of the pixels corresponding to the doubled direction angle are synthesized to obtain a synthesized vector. The formula for calculating the synthesized vector is: In the above formula, and These represent the components of the composite vector in the x and y directions, respectively. S2133. Calculate the direction angle of the composite vector and divide it by 2 to obtain the true direction angle of the composite vector. S2134. Calculate the average direction vector based on the true direction angle of the composite vector. The calculation formula is as follows: In the above formula, Represents the average direction vector. This represents the true direction angle of the composite vector.

[0039] S214. Calculate the mean value of the projection of the gray-level gradient vector of each pixel within the rectangular unit onto the average direction vector to obtain the gradient consistency index. The calculation formula is as follows: In the above formula, This represents the gradient consistency index of the rectangular cell located in the m-th row and n-th column after meshing. This represents the average direction vector of the rectangular cell located in the m-th row and n-th column after meshing. This represents the set of all pixels located in the rectangular cell at row m and column n after meshing. Indicates the size of the set. The gradient consistency index represents the L2 norm of the vector. It is used to reflect the extent to which the gray-level gradient direction of all pixels in the rectangular unit is consistent with the main direction. If the value of the gradient consistency index corresponding to the rectangular unit is closer to 1, it means that the directionality of the texture in the rectangular unit is stronger and more consistent, that is, the higher the possibility of directional wrinkles.

[0040] S215. Gradient consistency alone is insufficient to confirm wrinkles, as a scratch or strip-shaped impurity may also have directional consistency. Considering that another core feature of wrinkles is the periodic alternation of light and dark in the direction perpendicular to their direction, i.e., grayscale fluctuates in a wave-like manner, multiple parallel grayscale profile lines, all perpendicular to the grayscale gradient vector, are established on the plane where the rectangular unit is located. The grayscale values ​​corresponding to each pixel on the grayscale profile line are obtained. The grayscale values ​​of each pixel are arranged sequentially from one end of the grayscale profile line to the other to obtain a grayscale sequence. By analyzing the changes in grayscale values ​​in the grayscale sequence, the fluctuations in the grayscale values ​​of pixels in continuous space can be obtained.

[0041] S216. Based on the total number of peaks and troughs corresponding to all grayscale sequences in the rectangular unit, calculate the average of the total number of peaks and troughs contained in each grayscale profile line to obtain the grayscale volatility index. The calculation formula is as follows: In the above formula, This represents the grayscale fluctuation index of the rectangular cell located in the m-th row and n-th column after gridding. This represents the total number of peaks and troughs in the grayscale sequence corresponding to the k-th grayscale profile line. A grayscale profile line. The larger the value of the grayscale fluctuation index, the more frequent the grayscale changes perpendicular to the texture direction within the rectangular unit, the stronger the periodicity, and the higher the possibility of wrinkles.

[0042] S217. True wrinkles should simultaneously satisfy two conditions: consistent texture direction and periodic fluctuation of grayscale value perpendicular to the texture direction. A high gradient consistency index alone may indicate a scratch, while a high grayscale fluctuation index alone may indicate random noise. Therefore, a comprehensive evaluation based on both conditions is required. That is, the local wrinkling index is calculated based on the gradient consistency index and grayscale fluctuation index of each rectangular unit. The calculation formula is as follows: In the above formula, This represents the local wrinkling index of the rectangular cell located in the m-th row and n-th column after meshing. The larger the value, the higher the probability and severity of wrinkling of the rectangular cell. Used to ensure that the logarithm is positive.

[0043] S218. The need for tension control is difficult to achieve based on the local wrinkling index data of a large number of individual rectangular units. For tension control, a more intuitive display of the overall wrinkling along the width of the cling film is required. Therefore, the cling film is divided into multiple non-overlapping statistical units along its forward movement direction. Figure 2 The area between any two adjacent red lines in the diagram. Figure 2 The system is divided into 5 statistical units, and the ratio of the number of rectangular units with a local wrinkling index greater than the preset wrinkling index threshold to the total number of rectangular units is calculated in turn to obtain the instantaneous wrinkling index of each statistical unit.

[0044] S219. Considering that under stable production conditions, slight film vibration, instantaneous changes in illumination, or minor fluctuations in the detection algorithm may cause sudden changes in the calculation results, directly using this data as the basis for subsequent calculations may cause fluctuations in the pressure roller parameters. Therefore, the average value of the instantaneous wrinkling index of each statistical unit at each moment within a certain period of time is calculated. For example, the average value of the instantaneous wrinkling index of each statistical unit at each moment within 1 minute from the current moment is calculated to obtain the first quantitative characteristic value of the statistical unit at the current moment, thereby achieving the purpose of smoothing noise.

[0045] When calculating the first quantitative characteristic value, it is also necessary to simultaneously calculate the second quantitative characteristic value, which reflects the degree of film deviation, and the third quantitative characteristic value, which reflects the trend of film deviation. Deviation of the cling film can lead to both a poorer winding effect and wrinkles. The specific calculation steps are as follows: S221. To measure the degree of plastic wrap deviation, it is necessary to compare the position of the plastic wrap in the standardized image with its position when it is not deviationed in real time. Therefore, it is necessary to first define the position when it is not deviationed. After the equipment is installed and debugged and is in an ideally aligned state, the position of the two edges of the film at this time is recorded by the vision system and fitted to two straight lines, namely the theoretical baseline. Figure 2 The two blue lines in the image are permanently stored in the system as the absolute spatial reference for all subsequent real-time images to calculate deviation. Therefore, two pre-calibrated parallel theoretical baselines are established on the standardized image to reflect the position of the edge of the plastic wrap. The purpose of setting two parallel theoretical baselines instead of a single center line is to simultaneously monitor the overall width of the film and the position of the center line. If only the center line is monitored, when the film is slightly twisted, such as when one side of the plastic wrap is in front and the other side is behind, its center line may remain unchanged, but it has actually deviated.

[0046] S222. Edge detection is performed on the standardized image in real time using edge detection operators such as Canny and Sobel to identify the actual pixel points on both sides of the plastic wrap edge.

[0047] S223. Fit the actual pixel points on both sides of the plastic wrap to obtain the real-time edge lines of both sides of the plastic wrap.

[0048] S224. To comprehensively assess the degree of deviation of the plastic wrap, sampling measurements need to be taken at multiple locations along the direction of the plastic wrap's movement to avoid the randomness of single-point sampling. Multiple measurement lines perpendicular to and parallel to the two theoretical baselines should be established on the standardized image. Figure 2 The green line in the figure is used to obtain the intersection points of each measurement line with the theoretical baseline and the real-time edge line, and these intersection points are marked as theoretical intersection points and real-time intersection points respectively. The total number of theoretical intersection points and real-time intersection points on each straight line is 4, that is, each side edge of the plastic wrap corresponds to one theoretical intersection point and one real-time intersection point.

[0049] S225. Based on the coordinates of the theoretical intersection point and the real-time intersection point, calculate the Euclidean distance between the theoretical intersection point and the real-time intersection point on the same side edge of the plastic wrap on each measurement line in turn to obtain the deviation on one side. At the same time, in order for the subsequent control algorithm to clearly define the direction of correction, this distance must be assigned directionality. Therefore, the directionality can be assigned as follows: when the real-time intersection point on the same side edge of the plastic wrap is located on one side of the corresponding theoretical baseline (such as the left side), the Euclidean distance is marked as positive; when the real-time intersection point is located on the other side of the theoretical baseline, the Euclidean distance is marked as negative.

[0050] S226. The unilateral deviation reflects the local positional deviation of a single edge at the location of the measurement line. However, tension control is more concerned with the overall lateral positional change of the film. For example, when the entire cling film shifts to one side, or when there is a deviation in the width dimension of the film processing, simply using the data from one side to represent the whole may be inaccurate. Therefore, calculating the average of the deviations on both sides can effectively condense the overall position of the cling film at the measurement line into a scalar, that is, calculating the average of the two unilateral deviations corresponding to the measurement line to obtain the local deviation.

[0051] S227. Calculate the average value of the local deviation corresponding to each measurement line to obtain the instantaneous deviation, and mark it as the second quantization feature value, so as to reflect the degree of deviation of the plastic wrap as a whole, avoid the error of single-point sampling, and arrange the instantaneous deviation calculated at each time in chronological order to obtain the instantaneous deviation sequence.

[0052] S228. Set a statistical window for a period of time from the current moment, and perform linear regression analysis on the data of the instantaneous deviation sequence within the statistical window. Calculate the regression slope to obtain the third quantitative feature value. Linear regression analysis is a standard mathematical tool for extracting trends from time series. The second quantitative feature value reflects the current degree of deviation of the plastic wrap, while the regression slope reflects the rate of change of deviation. When the regression slope is large, even if the second quantitative feature value is not large, it is necessary to intervene in advance to prevent quality problems from occurring.

[0053] S3. Input the first, second, and third quantized feature values, along with the real-time process parameters of the winding process, into a pre-established tension mapping model that characterizes the degree of film wrinkling, the degree of film deviation, and the relationship between process parameters and pressure roller adjustment. The tension mapping model is constructed based on a reinforcement learning model. The pressure roller adjustment includes the pressure adjustment, tilt angle adjustment, and lateral displacement adjustment. Traditional PID control relies on precise models and fixed targets, such as constant tension, which makes it difficult to handle problems such as eliminating visually visible wrinkles and deviation as proposed in this embodiment. Reinforcement learning, by allowing the agent to learn the optimal control strategy through continuous interaction with the environment, provides a solution to the problems proposed in this embodiment. The specific steps are as follows: A state space is constructed, which includes the second and third quantitative feature values ​​at the current moment, the first quantitative feature value corresponding to each statistical unit, the current pressure of the pressure roller, the current tilt angle of the pressure roller, the current lateral displacement of the pressure roller, and the real-time process parameters of the winding process. The real-time process parameters generally include the linear speed of the plastic wrap winding and the real-time roll diameter.

[0054] The motion space is constructed, which includes three adjustment amounts of the pressure roller: pressure adjustment, tilt angle adjustment, and lateral displacement adjustment.

[0055] A reward function is constructed based on the first, second, and third quantified characteristic values ​​at the next and current times. Specifically, the reward function is a weighted sum of a wrinkle improvement reward, a deviation correction reward, a tension stability reward, and a safety constraint penalty. The wrinkle improvement reward is calculated by summing the reductions in the first quantified characteristic values ​​for each statistical unit compared to the current time, and is positively correlated with this sum. The deviation correction reward is calculated by summing the reductions in the absolute values ​​of the second and third quantified characteristic values ​​compared to the current time, and is positively correlated with these reductions. The tension stability reward is calculated by normalizing the current pressure adjustment, roller tilt angle adjustment, roller lateral displacement adjustment, and the maximum allowable adjustment range of each parameter, calculating the sum of squares of the normalized parameters, and is negatively correlated with this sum. The safety constraint penalty is set to a value much higher than the preset values ​​of the wrinkle improvement reward, deviation correction reward, and tension stability reward when the tension of the adjusted plastic wrap exceeds its allowable safety tension threshold. This increases the severity of the safety constraint penalty, prioritizing safety above all else. Otherwise, the penalty is 0.

[0056] Finally, based on the constructed state space, action space, and reward function, a deep reinforcement learning algorithm, such as deep deterministic policy gradient, is used to train the tension mapping model. The trained tension mapping model is then deployed on the production line to achieve real-time calculation of the pressure roller adjustment amount.

[0057] S4. After obtaining the pressure roller adjustment amount, the servo motor and other actuators will adjust the pressure roller parameters according to the pressure roller adjustment amount. For example, the pressure adjustment amount can be achieved by controlling the pressure roller to move up and down in the direction perpendicular to the plastic wrap. The pressure change is determined by the value fed back by the pressure sensor. The pressure increases when the roller is close to the plastic wrap and decreases when it is away from the plastic wrap. The tilt angle of the pressure roller can be adjusted by adjusting the distance of one end of the pressure roller in the direction perpendicular to the plastic wrap. The lateral displacement adjustment amount can be adjusted by controlling the pressure roller to move horizontally in the width direction of the plastic wrap. The direction can be defined as positive and negative. The direction definition should be consistent with the definition method of the single-sided deviation amount mentioned above.

[0058] The present invention also provides a tension control system for the plastic wrap winding process, including a processor and a memory, wherein the memory is used to store a computer program, and the computer program is executed by the processor to implement the tension control method.

[0059] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Since the above embodiments are substantially similar to the method embodiments, their descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0061] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A tension control method for the plastic wrap winding process, applied to the pressure roller for tension control in the plastic wrap winding process, characterized in that, The tension control method includes the following steps: S1. Collect the original film image sequence in the flat state before the plastic wrap is rolled up, and perform preprocessing to obtain a standardized image sequence containing multiple standardized images collected periodically. S2. Based on multiple standardized image sequences, calculate the first quantitative feature value in each standardized image that reflects the degree of wrinkling on the membrane surface, the second quantitative feature value that reflects the degree of membrane surface deviation, and the third quantitative feature value that reflects the trend of membrane surface deviation. S3. Input the first quantitative feature value, the second quantitative feature value, the third quantitative feature value and the real-time process parameters of the winding process into a pre-established tension mapping model for characterizing the degree of film wrinkling, the degree of film deviation, and the relationship between process parameters and pressure roller adjustment amount. The tension mapping model is constructed based on a reinforcement learning model. The pressure roller adjustment amount includes the pressure adjustment amount, tilt angle adjustment amount and lateral displacement adjustment amount of the pressure roller. S4. Adjust the pressure roller parameters according to the pressure roller adjustment amount.

2. The tension control method according to claim 1, characterized in that, In step S1, the specific steps for acquiring the original film image sequence in its flattened state before rolling are as follows: S111. Set the light source to shine on the surface of the plastic wrap, and the angle between the incident direction of the light and the normal of the surface of the plastic wrap is greater than 75 degrees. S112. A light-absorbing background is provided on the back of the plastic wrap, wherein the light-absorbing background is a black rough surface material with a spectral reflectance of less than 5%. S113. An image acquisition device is set up to capture images of the plastic wrap in a flat state before it is rolled up, and the angle between the shooting direction of the image acquisition device and the illumination direction of the light source is in the range of [15°, 45°]. S114. Images of the plastic wrap at various times are acquired in real time using an image acquisition device and arranged in chronological order to obtain the original film image sequence.

3. The tension control method according to claim 1, characterized in that, In step S1, the specific steps of the preprocessing are as follows: S121. Filter each original thin film image in the original thin film image sequence to obtain a filtered image; S122. Construct a background function model to characterize the ideal background grayscale value of each pixel in the filtered image. Its expression is: In the above formula, Indicates the coordinates in the filtered image as The ideal background grayscale value for the pixel at point n, where n is the order of the polynomial function. For coefficients; S123. Grid the filtered image and calculate the standard deviation of the gray values ​​of each pixel in each grid in turn; S124. Select multiple grids with the smallest standard deviation as candidate grids; S125. Calculate the average gray value of all pixels in each candidate grid in turn, and select the pixel corresponding to the gray value that is closest to the average value as the candidate pixel corresponding to the candidate grid. S126. By fitting the background function model using the least squares method, the sum of squared errors between the gray values ​​of the background function model and the true gray values ​​at all candidate pixels is minimized in order to solve for the coefficients and obtain the background function model. S127. Subtract the background function model from the filtered image and adjust the grayscale values ​​to obtain the corrected uniform illumination image. The expression is: In the above formula, Represents coordinates in a uniformly lit image The pixel value at that pixel. Indicates the coordinates in the filtered image as The pixel value at the pixel point, and C represents the adjustment coefficient used to adjust the pixel value of each pixel in the uniformly illuminated image to a suitable level; S128. Linearly contrast stretch the uniformly illuminated image to make its grayscale value range reach the preset range to obtain a standardized image, and a standardized image sequence composed of multiple standardized images arranged in time sequence.

4. The tension control method according to claim 1, characterized in that, In step S2, the specific steps for calculating the first quantized feature value are as follows: S211. The standardized image is meshed to obtain multiple non-overlapping rectangular units; S212. Calculate the gray-level gradient vector of each pixel within each rectangular unit. The expression is as follows: In the above formula, Represents pixels The gray gradient vector at that point, and Representing pixels The gradient components of the gray-level gradient vector at a given location in the x and y directions; S213. Normalize the gray-level gradient vector of each pixel in the rectangular unit, and calculate the average direction vector that represents the main direction of the gray-level gradient vector corresponding to each pixel in the rectangular unit. S214. Calculate the mean value of the projection of the gray-level gradient vector of each pixel within the rectangular unit onto the average direction vector to obtain the gradient consistency index. The calculation formula is as follows: In the above formula, This represents the gradient consistency index of the rectangular cell located in the m-th row and n-th column after meshing. This represents the average direction vector of the rectangular cell located in the m-th row and n-th column after meshing. This represents the set of all pixels located in the rectangular cell at row m and column n after meshing. Indicates the size of the set. The L2 norm of a vector; S215. Establish multiple parallel grayscale profile lines on the plane where the rectangular unit is located, all of which are perpendicular to the grayscale gradient vector, and obtain the grayscale value corresponding to each pixel on the grayscale profile line to obtain the grayscale sequence. S216. Based on the total number of peaks and troughs corresponding to all grayscale sequences in the rectangular unit, calculate the average of the total number of peaks and troughs contained in each grayscale profile line to obtain the grayscale volatility index. The calculation formula is as follows: In the above formula, This represents the grayscale fluctuation index of the rectangular cell located in the m-th row and n-th column after gridding. This represents the total number of peaks and troughs in the grayscale sequence corresponding to the k-th grayscale profile line. grayscale profile lines; S217. Calculate the local wrinkling index of each rectangular unit based on its gradient consistency index and grayscale fluctuation index. The calculation formula is as follows: In the above formula, This represents the local wrinkling index of the rectangular cell located in the m-th row and n-th column after meshing; S218. Divide the plastic wrap into multiple non-overlapping statistical units along the direction of movement of the plastic wrap, and calculate the ratio of the number of rectangular units in each statistical unit whose local wrinkling index is greater than the preset wrinkling index threshold to the total number of rectangular units in order to obtain the instantaneous wrinkling index of each statistical unit. S219. Calculate the average value of the instantaneous wrinkling index of each statistical unit within a certain period of time from the current time, so as to obtain the first quantitative characteristic value of the statistical unit at the current time.

5. The tension control method according to claim 4, characterized in that, In step S213, the calculation steps for the average direction vector are as follows: S2131. Calculate and double the orientation angle of each pixel within the rectangular unit based on its grayscale gradient vector. The calculation formula is as follows: In the above formula, Represents the coordinates within a rectangular cell The direction angle of the grayscale gradient vector corresponding to the pixel at that location. It is the angle after doubling the direction angle. and Representing pixels The gradient components of the gray-level gradient vector at a given location in the x and y directions; S2132. The normalized grayscale gradient vectors of the pixels corresponding to the doubled direction angle are synthesized to obtain a synthesized vector. The formula for calculating the synthesized vector is: In the above formula, and These represent the components of the composite vector in the x and y directions, respectively. S2133. Calculate the direction angle of the composite vector and divide it by 2 to obtain the true direction angle of the composite vector. S2134. Calculate the average direction vector based on the true direction angle of the composite vector.

6. The tension control method according to claim 1, characterized in that, The specific steps for calculating the second and third quantization feature values ​​are as follows: S221. Establish two pre-calibrated parallel theoretical baselines on the standardized image to reflect the position of the edge of the plastic wrap. S222. Perform edge detection on the standardized image in real time to identify the actual pixel points on both sides of the plastic wrap. S223. Fit the actual pixel points on both sides of the plastic wrap to obtain the real-time edge lines on both sides of the plastic wrap. S224. Establish multiple measurement lines on the standardized image that are perpendicular to and parallel to the two theoretical baselines, and obtain the intersection points of each measurement line with the theoretical baseline and the real-time edge line, and mark them as theoretical intersection points and real-time intersection points respectively. S225. Based on the coordinates of the theoretical intersection point and the real-time intersection point, calculate the Euclidean distance between the theoretical intersection point and the real-time intersection point on the same side edge of the plastic wrap on each measurement line in turn to obtain the deviation on one side. When the real-time intersection point is located on one side of the theoretical baseline, mark the Euclidean distance as positive and when the real-time intersection point is located on the other side of the theoretical baseline, mark the Euclidean distance as negative. S226. Calculate the average value of the deviation on the two sides corresponding to the measurement line to obtain the local deviation. S227. Calculate the average value of the local deviation corresponding to each measurement line to obtain the instantaneous deviation, and mark it as the second quantization feature value. Arrange the instantaneous deviation calculated at each time point in chronological order to obtain the instantaneous deviation sequence. S228. Set a statistical window for a period of time from the current time, and perform linear regression analysis on the data of the instantaneous deviation sequence within the statistical window to calculate its regression slope and obtain the third quantitative feature value.

7. The tension control method according to claim 1, characterized in that, In step S3, the tension mapping model is constructed based on a reinforcement learning model, and the specific steps are as follows: A state space is constructed, which includes the second and third quantized feature values ​​at the current moment, the first quantized feature values ​​corresponding to each statistical unit, the current pressure of the pressure roller, the current tilt angle of the pressure roller, the current lateral displacement of the pressure roller, and the real-time process parameters of the winding process. Construct an action space, which includes three pressure roller adjustment amounts: pressure adjustment amount, tilt angle adjustment amount, and lateral displacement adjustment amount. Construct a reward function, which is calculated based on the first quantized feature value, the second quantized feature value, and the third quantized feature value at the next time step and the current time step; Based on the state space, action space, and reward function, a deep reinforcement learning algorithm is used to train the tension mapping model.

8. The tension control method according to claim 7, characterized in that, The reward function is a weighted sum of the wrinkle improvement reward, deviation correction reward, tension stability reward, and safety constraint penalty. The wrinkle improvement reward is calculated by summing the reduction of the first quantitative feature value corresponding to each statistical unit compared to the current time, and is positively correlated with the sum of the reductions. The deviation correction reward is calculated by comparing the absolute values ​​of the second and third quantitative feature values ​​at the next time step with the current time step, and is positively correlated with the absolute values ​​of the second and third quantitative feature values. The tension stability bonus is calculated by normalizing the current pressure adjustment, the tilt angle adjustment of the pressure roller, the lateral displacement adjustment of the pressure roller, and the maximum allowable adjustment range of each parameter, and then calculating the sum of squares of the normalized parameters and negatively correlated with this sum of squares. The safety constraint penalty is set to a value that is much greater than the preset values ​​of the wrinkle improvement reward, deviation correction reward, and tension stability reward when the tension of the adjusted plastic wrap is greater than its allowable safety tension threshold; otherwise, it is 0.

9. The tension control method according to claim 5, characterized in that, In step S2134, the formula for calculating the average direction vector is: In the above formula, This represents the average direction vector. This represents the true direction angle of the composite vector.

10. A system for implementing the tension control method according to any one of claims 1-9, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, which, when executed by the processor, implements the tension control method as described in any one of claims 1-9.

Images